Multimodal Reasoning on CogReasonBench 1.0 (test)
3.985MM Intent ScoreCogVLM
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| CogVLMTraining Strategy=RFT2026.05 | 3.985 | 4.449 | 4.599 | 4.959 | 4.473 | |
| CogVLMTraining Strategy=SFT2026.05 | 3.725 | 4.445 | 4.266 | 4.955 | 4.343 | |
| Qwen3-VL-8B-Thinking2026.05 | 2.67 | 3.824 | 3.829 | 4.727 | 3.752 | |
| Qwen3-VL-8B-Instruct2026.05 | 2.48 | 4.045 | 3.905 | 4.42 | 3.712 |